Abstract
Letter identification is a canonical topic in the study of human visual pattern identification. It serves as a natural example of identification of a finite set of well-learned simple spatial patterns, and also underlies most tests of visual acuity. As part of our effort to develop a model of image identification (Watson & Ahumada, 2008, doi:10.1167/8.4.17; Watson & Ahumada, 2012, doi:10.1167/12.10.19) we have reviewed the published data for contrast thresholds for letter identification as a function of size. Eight studies, examining letter sizes from 0.04 to 60 degrees and employing a variety of stimulus materials and conditions, agree that sensitivity increases rapidly with size from the acuity limit, but then saturates, and may decline slightly, for letter sizes above about 1 degree. None of these studies has offered a comprehensive account of this pattern of results. We have explored the performance of a template model limited by optics, eccentricity-dependent retinal filtering and sampling, neural noise, and an efficiency that varies with pattern size. Optical limits were derived as the mean MTF from wavefront aberrations of 200 normal observers (Thibos et al., 2002, doi:10.1364/JOSAA.19.002329). Retinal limits were derived from a formula for human midget retinal ganglion cell densities developed by Drasdo et al. (2007, doi:10.1016/j.visres.2007.01.007) and extended by us to two dimensions. Anisoplanatic spatial filtering used a generalization of the method of Perry and Geisler (2002, doi:10.1117/12.469554). Efficiency values were derived from Pelli et al. (2006, doi:10.1016/j.visres.2006.04.023). Noise value was derived from template model fits to ModelFest Gabor contrast thresholds (Watson & Ahumada, 2005, doi:10.1167/5.9.6). With no free parameters, the model provides a good description of the ensemble data. We find that all model components are required to account for the pattern of results.
Meeting abstract presented at VSS 2013